ai-engineering

Production AI agents require integrated system prompt architecture, tool-call schemas, and context engineering to prevent hallucinations. This Skill builds reliable agent systems by enforcing capability allowlists, guardrail hierarchies, and confirmation policies for safe tool execution.

Eval regression gates for LLM behavior are set by defining eval contracts, datasets, and CI gate cascades. This Skill configures hard_fail and soft_fail thresholds to ensure safe iteration and prevent regression during agent deployment.

Context engineering improves RAG assembly by routing context via tagger patterns and applying retrieval compaction with diversity. This Skill debugs wrong-context failures and assembles retrieval architectures to ground LLM responses reliably.

Memory systems for multi-agent orchestration are managed by separating working, run, and cross-run memory. This Skill implements reflection loops and connects memory to fine-tuning pipelines, applying manager and handoff patterns for production governance.

AI agents select wrong tools when tool schemas and categories lack strict confirmation and persistence policies. This Skill debugs wrong-tool selection by defining robust tool call architecture and guardrail hierarchies to constrain agent decisions.

System prompt design for fine-tuning pipelines is supported by generating capability allowlists and SPL-style instruction learning patterns. This Skill connects memory systems and reflection loops directly into fine-tuning data infrastructure for reliable behavior.